[
https://issues.apache.org/jira/browse/PDFBOX-1912?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13911811#comment-13911811
]
John Hewson commented on PDFBOX-1912:
-------------------------------------
Asked on the mailing list:
{quote}
why shouldn't we directly do OCR on those
PDFs without getting output from PDFBox?
{quote}
PDFBox can generate images (PDFToImage) and can extract text (PDFToText). The
goal of
this project is to enhance PDFToText so that it can use OCR to extract text
from areas of the
document where the text is embedded as an image. Such PDF files are typically
generated by
scanners or fax machines. There is also another case where OCR is useful: some
fonts embedded
in PDF files contain the wrong encoding, so when text is extracted with
PDFToText the result is
nonsense but when drawn with PDFToImage we see the correct letters.
Instead of:
{{PDF => Image => OCR => Text }}
We want to do:
{{PDF => (Many images for words + location data => OCR) => Text}}
> Optical Character Recognition (OCR)
> -----------------------------------
>
> Key: PDFBOX-1912
> URL: https://issues.apache.org/jira/browse/PDFBOX-1912
> Project: PDFBox
> Issue Type: Wish
> Components: Text extraction
> Affects Versions: 2.0.0
> Environment: JDK 6, C/C++
> Reporter: John Hewson
> Assignee: John Hewson
> Labels: gsoc2014
>
> Brief explanation: The PDFBox library is widely used to extract text from PDF
> files. However, many PDF files embed text in a malformed manner which renders
> text extraction useless. There has recently been interest in extracting
> governmental data from PDF files, the PDF Liberation commons being a notable
> example, see https://github.com/pdfliberation for more details.
> Many end-users of PDFBox have been making use of OCR tools such as Google's
> Tesseract https://code.google.com/p/tesseract-ocr/ which are run on the final
> image generated by PDFBox. We think that by adding a more integrated OCR API
> to PDFBox it will be possible to do a better job. PDFBox often has access to
> encoding and positioning information for individual glyphs. Even when their
> extracted text is meaningless, a character-by-character, or line-by-line OCR
> could be more accurate. PDFBox also has information such as image orientation
> which could allow it to better perform OCR on pages such as embedded
> landscape tables.
> There are existing JNI bindings for Tesseract available at
> https://code.google.com/p/tesseract-android-tools/
> Expected results: To extend PDF box with an API which allows external OCR
> tools to be plugged-in, and an implementation of a Tesseract plug-in using
> either JNI or the command line via Process.exec.
> Knowledge Prerequisite: Java, JNI (C/C++)
> Mentor: John Hewson
> PMC Note: Tesseract is under the Apache License 2.0
> To learn more about PDFBox, please visit http://pdfbox.apache.org/
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
